rea-video.Video Breakdown Assistant
中文Get started
Production guides

Why your video prompt does not match the reference

A mismatch is easier to fix when you name it. Compare one shot at a time and change the part responsible for the difference.

Updated

Case evidence: Read the source breakdown report · View case frames and shot timings

Based on a published tool report. Production classifications are analytical judgments; suggested steps are not the original project or a verified recreation.

The scene is right, but the motion is wrong

Write the start state, one visible action and the end state. Keep camera movement on a separate line. If you describe only a still frame, the model has little guidance about what should change. In the unboxing example, “headphone box on a carpet” omits the sponge motion that defines the shot.

The character changes between cuts

The dialogue report separates character shots from interface and title shots. Keep a consistent character description and permitted reference assets across character shots. Test continuity before assembling the full sequence. A text prompt alone cannot guarantee the same face, costume or pose across independent generations.

The captions or interface look wrong

In the dialogue case, shots 2 and 3 include text overlays in addition to character footage. Treat these as separate production tasks: generate or shoot the base image, then typeset the exact words. For a real interface, use a recording or rebuild the required layout rather than asking a video model to invent readable controls.

The clip looks similar but feels too slow

Use the reference cut points: the dialogue report places shot 2 at 00:01.375–00:02.833 and shot 3 at 00:02.833–00:03.875. Compare action timing and cut timing separately. Trim usable footage, then align speech and effects. Adding “fast-paced” to every prompt is not a substitute for setting edit points.

Keep a small test log

Record the input reference, prompt, model settings you actually used, result and one next change. Score framing, motion, identity, text and timing separately. If a change helps one dimension but harms another, keep the evidence instead of declaring the whole prompt successful. The report cited here is source analysis; these troubleshooting steps have not been benchmarked across models.

A reusable analysis template

Reference shot / time range:
Observed mismatch:
Layer: picture / camera / text / audio / edit
One change to try:
Settings and output saved:
Result: improved / unchanged / worse
Next decision:

Practice with a real report

Use shots 2 and 3 to practice separating character continuity, overlay typography and cut timing.

Open the case and download its report →

Continue your workflow